---
title: "GPT‑5.4 Benchmarks - RankedAGI"
description: "Detailed benchmark and metadata record for GPT‑5.4."
source: "https://rankedagi.com/models/gpt-5-4"
---

# GPT‑5.4

| Field | Value |
| --- | --- |
| Organization | OpenAI |
| License | Proprietary |
| Type | Thinking |
| Version | Latest |
| Released | 2026-03-05 |
| Context window | 1M |
| Knowledge cutoff | 2025-08-31 |
| Input cost per million tokens | $2.5 |
| Output cost per million tokens | $15 |
| Cached input cost per million tokens | $0.25 |
| Model ID | gpt-5.4 |
| Last updated | 2026-05-28T08:57:27.471Z |

## Benchmarks

| Benchmark | Category | Value | Description | Source |
| --- | --- | --- | --- | --- |
| RankedAGI Coding | coding | 81.2% | RankedAGI Coding Score |  |
| SWEBench Pro | coding | 57.7% | Diverse Agentic Coding |  |
| Terminal-Bench 2.0 | coding | 75.1% | Agentic Terminal Coding | https://www.tbench.ai/leaderboard/terminal-bench/2.0 |
| RankedAGI Agentic | Agents | 75.6% | RankedAGI Agentic Score |  |
| BrowseComp | Agents | 82.7% | A benchmark for browsing agents | https://openai.com/index/browsecomp/ |
| Code Arena | coding | 1457 | Code Arena Elo Score | https://arena.ai/leaderboard/code |
| Livebench Coding 26.1 | coding | 77.5% | Livebench Coding 26.01.08 | https://livebench.ai/#/?Coding=a&sort=Coding+Average&variants=true&highunseenbias=true |
| DeepSWE | coding | 56.0% | Measuring frontier coding agents on original, long-horizon engineering tasks | https://deepswe.datacurve.ai/ |
| LiveBench Agentic Coding 26.1 | coding | 70.0% |  | https://livebench.ai/#/?Agentic+Coding=a&sort=Agentic+Coding+Average&variants=true&highunseenbias=true |
| OSWorld-Verified | Agents | 75.0% |  |  |
| SvelteBench | coding | 95.6% | SvelteBench - Benchmark for Svelte | https://khromov.github.io/svelte-bench/benchmark-results-merged.html |
| CyberGym | safety | 79.0% |  |  |
| Codeforces ELO Score | coding | 3168 | Codeforces ELO Score |  |
| RankedAGI Reasoning | reasoning | 77.6% | RankedAGI Reasoning Score |  |
| Humanity's Last Exam | reasoning | 39.8% | Multidisciplinary Reasoning (no tools) |  |
| Humanity's Last Exam w/ Tools | reasoning | 52.1% | Multidisciplinary Reasoning (with tools) |  |
| GPQA Diamond | reasoning | 92.8% | Generalized Prefix Question Answering Score (Reasoning) PhD Level Reasoning |  |
| Text Arena | general | 1482 | ChatArena (LMSYS) ELO Score | https://arena.ai/leaderboard/text |
| AIME 2026 | math | 95.2% | AIME 2026 Competition Math | https://matharena.ai/?comp=aime--aime_2026&view=problem |
| Vending Bench 2 | Agents | $6,144.18 | Benchmark for measuring AI model performance on running a business over long time horizons. Models are tasked with running a simulated vending machine business over a year and scored on their bank account balance at the end. | https://andonlabs.com/evals/vending-bench-2 |
| NYT Connections Extended | reasoning | 94.0% | NYT Connections Extended Version | https://github.com/lechmazur/nyt-connections/ |
| MMLU Pro | knowledge | 87.5% | Graduate-level visual problem-solving |  |
| MMMU-Pro | imaging | 81.2% | Multimodal Understanding |  |
| MMMU Pro w/Tools | imaging | 82.1% |  |  |
| RankedAGI Math | math | 76.5% | RankedAGI Math Score |  |
| RankedAGI Overall | general | 69.9% | Overall RankedAGI score |  |
| ARC AGI 2 | reasoning | 73.3% | Abstract Reasoning Puzzles (Public) |  |
| LiveCodeBench Pro | coding | 87.5% | Competitive Coding |  |
| 𝜏²-Bench Telecom | Agents | 92.8% | Agentic Tool Use |  |
| HealthBench Hard | Health | 40.1% | Open-ended Health Queries |  |
| MedXpertQA Text | Health | 59.6% | Medical Multiple Choice |  |
| MedXpertQA (MM) | Health | 77.1% | Medical Multiple Choice |  |
| DeepSearch QA | Agents | 73.6% | Agentic Search |  |
| GDPval-AA Elo | Agents | 1674 | Office Tasks (Artificial Analysis) |  |
| ZeroBench | general | 41.0% | Multi-step Visual Reasoning |  |
| MCP Atlas | Agents | 70.6% | Scaled Tool Use |  |
| Finance Agent | Agents | 56.0% | Agentic Financial Analysis |  |
| CharXiv Reasoning | Multimodal | 82.8% | Figure Understanding |  |
| Code DesignArena | Design | 1282 |  | https://www.designarena.ai/leaderboard/code |
| Toolathlon | Agents | 54.6% | Benchmark to assess language agents' general tool use in realistic environments. | https://toolathlon.xyz/introduction |
| Tau3 Bench | Agents | 72.9% |  | https://sierra.ai/resources/research/tau-3-bench |

## Model Links

| Title | URL |
| --- | --- |
| Release | https://openai.com/index/introducing-gpt-5-4/ |